Selection of Morris trajectories for initial sensitivity analysis

The Morris method of initial sensitivity analysis of large models varies factors singly along a few trajectories. Its nominally uniform coverage of the factors may deviate greatly from uniform in such small samples. Campolongo et al. suggest selecting a subset of trajectories from a large set, to maximise their spread. Ways to reduce the computing demand of this procedure are examined, including alternative distance measures and two-stage selection. An example with more economical selection is used to investigate how the factors' marginal distributions deviate from uniform on trajectories selected by spread or by the number of factors whose values differ between trajectories.